Development of an apoE4-Aβ fibrillization assay
Recombinant human Aβ42 sodium hydroxide (NaOH) salt (rPeptide) was received following pre-treatment to ensure a consistent monomeric preparation, as described previously [40]. For NaOH pre-treatment of Aβ peptide, briefly, following recombinant protein expression and purification, Aβ42 peptides were dissolved in 2 mM NaOH, pH 10.5, and then sonicated and lyophilized. Upon receipt, the lyophilized peptide was reconstituted in ice-cold Dulbecco’s phosphate buffered saline (DPBS), pH 7.4, which avoids the solution passing through the isoelectric point of Aβ (pI = 5.5), which would induce aggregation [41]. The reconstituted Aβ42 stock solution was quickly aliquoted and snap-frozen in liquid nitrogen and then stored at −80°C until use. Great care was taken to ensure consistency and reproducibility across all experiments by using Aβ from a single batch, thawing and maintaining Aβ stock on ice until use, and never re-freezing the unused portion of thawed Aβ stock. For fibrillization experiments, Aβ42, recombinant human apoE4 (Sigma), and thioflavin T (ThT; Sigma) were combined at pre-determined concentrations in DPBS in a total volume of 40 µl in a 384-well µ-clear bottomed plate (Greiner). Plates were sealed to prevent evaporation and incubated at 37°C with constant rapid agitation and the fluorescence intensity of ThT at λex= 440 nm, λem= 490 nm was measured every 10 minutes (min) for up to 24 h using a Biotek Synergy HTX fluorescence plate reader and Gen5 v3.11 software (Biotek). Once the optimal concentrations of approximately 20 μM Aβ42, 1 nM apoE, and 15 μM ThT were determined, they were maintained throughout subsequent studies unless noted otherwise. For HTS assay validation, recombinant human Aβ40 (rPeptide), recombinant human scrambled Aβ42 (rPeptide), recombinant human apoE2 and apoE3 (Creative Biomart), recombinant human apolipoprotein A-I (apoA-I; Creative Biomart), human plasma-derived apoE (Sigma), or dimethyl sulfoxide (DMSO; Sigma) were included or substituted at the indicated concentrations. In assay optimization experiments, 3−8 wells were used per group and experiments were replicated one or two times, as indicated in the figure legends. When replicated twice, experiments were performed on different days and in different plates and the results of the two experiments were combined.
HTS of the NCC library
The NCC library was developed by the National Center for Advancing Translational Sciences (https://ncats.nih.gov/smr). Detailed information about these compounds is available using the NIH Chemical Genomics Center Pharmaceutical Collection browser [42]. The NCC library was received from Evotec, Inc., and contained each compound at 10 mM in DMSO which were aliquoted and stored at -80°C until use. To set up the exploratory drug screen, compounds were thawed, diluted in DMSO, and added at a concentration of 2 µM to Aβ42 (2 μM) in water, followed by the addition of apoE4 (20 nM) and the mixture was incubated at room temperature (rt) for 15 min. The mixture was then divided into three separate wells of a 96-well plate, ThT (8 μM) and glycine (30 mM) were added for a total volume of 125 μL per well, and the plate was incubated at rt for 10 min in the dark. The fluorescence intensity of ThT was then measured using the fluorescence plate reader. The 595 compounds were divided across numerous plates, and compounds on each plate were compared to control wells on the same plate that received Aβ42, apoE4, ThT, and DMSO. Unlike in the exploratory screen, the optimal concentrations of 20 μM Aβ42, 1 nM apoE, and 15 μM ThT were used in the HTS assay because this assay was developed and optimized for 384-well plates after the exploratory screen had already been completed. To set up the HTS assay, compounds were thawed, diluted in DMSO, and added to the Aβ42 in DPBS at final concentrations of 0.25, 2.5, and 25 µM in 5% DMSO/95% DPBS (v/v), followed immediately by the addition of apoE4 and ThT in a total volume of 40 μL per well. Plates were sealed to prevent evaporation and incubated at 37°C with constant shaking and the fluorescence intensity of ThT was measured every 10 min for 24 h using the fluorescence plate reader. The 87 compounds were divided across three separate plates, and compounds on each plate were compared to control wells on the same plate that received Aβ42, apoE4, ThT, and 5% DMSO. The criteria for hit identification were that the compound reduced ThT fluorescence by at least 30% at any concentration and that the effect was generally dose-dependent. For HTS, each compound was tested in 3−4 wells per concentration and experiments were replicated one or two times, as indicated in the figure legends.
Inhibition of Aβ alone and disaggregation of pre-formed fibrils
Each of the eight hit compounds was added to Aβ42 at 0.25, 2.5, and 25 µM in 5% DMSO/95% DPBS (v/v), followed immediately by the addition of ThT and measurement of fluorescence intensity every 10 min for 24 h. The area under the curve (AUC) of ThT fluorescence intensity was calculated and normalized to control wells receiving Aβ42, ThT, and 5% DMSO. To test compounds for disaggregation of pre-formed Aβ fibrils, Aβ42 and apoE4 were combined and incubated at 37°C for 24 h with constant shaking to induce fibrillization. Pre-formed Aβ fibrils were then divided into separate wells, and compounds were added in a final concentration of 5% DMSO and incubated at rt for 30 min with constant shaking. ThT was added to each well, the plates were incubated at rt for 15 min, and then fluorescence intensity was measured and normalized to control wells receiving only 5% DMSO. To test compounds for disaggregation of pre-formed tau fibrils, 2 μM recombinant human K18 tau peptide (Novus), comprising the microtubule binding domain of the 4R tau isoform, was combined with 2 μM heparin (Sigma) and 300 μM dithiothreitol (Invitrogen) in DPBS and incubated at 37°C for 24 h with constant shaking to induce fibrillization. Pre-formed tau fibrils were then divided into separate wells, and compounds were added in a final concentration of 5% DMSO and incubated at rt for 30 min with constant shaking. ThT (12.5 μM) was added to each well, the plates were incubated at rt for 15 min, and then fluorescence intensity was measured and normalized to control wells receiving only 5% DMSO.
Transmission electron microscopy (TEM)
Immediately following the measurement of ThT fluorescence intensity, pre-formed Aβ fibrils treated with individual hit compounds, or with DMSO as a control, were applied undiluted to Formvar/carbon-coated copper grids with 300 square mesh (Electron Microscopy Sciences) for 2 min. Grids were gently blotted on filter paper (Whatman) to remove excess fibrils, then washed twice in water and stained with 2% (w/v) uranyl acetate (Electron Microscopy Services) twice for 20 sec each, blotting on filter paper in between each step. Grids were air dried and imaged on a Tecnai G2 Spirit BioTwin microscope (FEI) at 80 kV with a side-mount digital camera (AMT Imaging). TEM images were processed and analyzed using Fiji version 2.1.0/1.53c.
Animals
5xFAD transgenic mice, which express the human APP gene harboring the Swedish (K670N/M671L), Florida (I716V), and London (V717I) familial AD mutations, and the human presenilin 1 (PSEN1) gene harboring the M146L and L286V familial AD mutations, from two separate transgenes, each driven by the murine Thy1 promoter, were originally developed on a mixed B6SJL background [43]. 5xFAD mice that had been backcrossed to a congenic C57BL/6J background (Jackson Labs # 034848-JAX) were received and maintained as a hemizygous line by breeding with C57BL/6J mice. TgF344-AD transgenic rats, which express the human APP gene harboring the Swedish (K670N/M671L) and the human PSEN1 gene with the Δ exon 9 mutation, both driven by the mouse prion protein promoter [44], were maintained on a Fischer 344 background. Mice and rats were treated in accordance with the Guide for the Care and Use of Laboratory Animals. All procedures were approved by the Institutional Animal Care and Use Committee of the University of Colorado.
5xFAD mouse primary neuron cell model
5xFAD mouse pups at postnatal day 1-2 were genotyped using primer probes and real-time polymerase chain reaction analysis of 1 mm tail snip samples. Brains from the mouse pups were then rapidly removed, cerebral cortices were isolated using a sterile razor blade, and tissue samples from multiple pups were pooled for experiments. Primary cultures of neurons were prepared using the Papain Dissociation System (Worthington) according to the manufacturer’s instructions. To prepare neuronal cultures, cortical tissue was dissociated in 20 U/ml papain under constant agitation at 37°C for 45 min. A single cell suspension was obtained by trituration, then papain was inactivated using ovomucoid protease inhibitor and cells were filtered through a 100 μm cell strainer and diluted in warm Neurobasal medium supplemented with Glutamax, B27 supplement, and penicillin/streptomycin (all from Gibco). Cells were seeded at 30,000 cells/cm2 in 96-well μ-clear bottomed plates (Ibidi) pre-coated with 10 μg/ml poly-D-lysine (Sigma). Neural cultures were maintained at 37°C in a humidified 5% CO2 chamber for 3 d, and then half of the culture medium was replaced with fresh medium also containing CultureOne supplement (Thermo Fisher), which reduces glial cell proliferation to favor neuronal culture. Every 3 d following exposure to Aβ and apoE4 and treatment with compounds, half of the culture medium was replaced and Aβ42, apoE4, and the test compounds in DMSO were added to maintain the initial concentrations. After 7 d in culture, half of the culture medium was replaced, and 100 nM Aβ42 and 1 nM apoE4 were added, followed by the addition of test compounds at 0.01, 0.1, or 1 μM in a final concentration of 0.5% (v/v) DMSO. At 9 days post-exposure (dpe), wells were fixed for immunocytochemistry and the conditioned media was collected for enzyme-linked immunosorbent assay (ELISA) analysis. We used the minimum number of mice to obtain sufficient numbers of cells to test all compounds in three wells per concentration. Cells from individual mice were pooled and used for all groups to remove the effect of biological variation and to allow us to use fewer mice.
Immunocytochemistry
At the pre-determined end points, the culture medium was removed, and the cells were washed once with DPBS, fixed in 4% (w/v) paraformaldehyde for 30 min, washed four times with DPBS, and stored at 4°C. The cells were permeabilized with 0.1% Triton X-100 in DPBS for 10 min and then blocked with 3% bovine serum albumin (BSA) in DPBS for 90 min and then incubated overnight at 4°C with primary antibodies in 3% BSA in DPBS. 5xFAD mouse cells were labeled with chicken anti-tau (PhosphoSolutions #1998-TAU, 1:1000) and mouse anti-Aβ (82E1, IBL #10323, 1:500) antibodies. TgF344-AD rat cells were labeled with chicken anti-tau, rabbit anti-Aβ (OC, Millipore #AB2286, 1:500), and mouse anti-pTau (AT8, Sigma #MN1020, 1:250) antibodies. Cells were washed and then incubated with Alexa Fluor Plus-conjugated secondary antibodies (Thermo Fisher, 1:500) for 45 min at rt in 3% BSA in DPBS. Cells were washed, and then nuclei were stained with 1 µg/ml Hoechst 33342 (Thermo Fisher) in DPBS for 10 min. The cells were then washed and imaged on an Olympus IX83 inverted fluorescence microscope. Images of entire wells were captured at 20X magnification and then analyzed using Cell Sens v1.12 software (Olympus).
Aβ ELISA analysis of conditioned media
Aβ concentration in conditioned medium from individual wells measured using the human Aβ42 ELISA kit (Thermo Fisher), following the manufacturer’s instructions. Two technical replicates were performed in the ELISA assay for each of three different wells per compound per concentration.
TgF344-AD rat primary neuron cell model
Brains from TgF344-AD transgenic rat pups at postnatal day 1 were removed, and cortices were isolated using a sterile razor blade. Primary cultures of neurons were prepared from cerebral cortices using the Papain Dissociation System (Worthington) according to the manufacturer’s instructions and were plated and cultured as described above for 5xFAD mouse neurons. Cells from individual rat pups were not pooled but were cultured in separate wells. After 7 d in culture, half of the culture medium was replaced, and 100 nM Aβ42 and 1 nM apoE4 were added, followed by the addition of test compounds at 1 μM in a final concentration of 0.5% (v/v) DMSO. Every 3 d thereafter, half of the culture medium was replaced and Aβ42, apoE4, and the test compounds in DMSO were added to maintain the initial concentrations. At 14 dpe, the cells were fixed for immunocytochemistry. We used the minimum number of rats to obtain sufficient numbers of cells to test all compounds in three wells per concentration. Cells from individual rat pups were not pooled in order to evaluate the drug effects on different biological replicates, although each drug and controls were tested on cells derived from the same rats.
NACC data analysis
The NACC uniform dataset v3 [45] was received on April 17, 2020 and contained standardized longitudinal clinical data on 42,661 subjects seen at Alzheimer’s Disease Research Centers (ADRCs) beginning in September, 2005 thru the March, 2020 data freeze. Subjects who had reported taking at least one of the eight hit compounds were identified by searching the ‘DRUGS’ column. Only subjects with at least two clinic visits and who had reported taking a medication prior to their final clinic visit were considered. Control groups of subjects taking antidepressant medications or antipsychotic medications were identified using the ‘NACCADEP’ or ‘NACCAPSY’ columns, respectively, with subjects who reported only taking imipramine or olanzapine being removed. The groups partially overlapped, as, for example, a subject may have reported using imipramine or olanzapine and then reported using a different antidepressant or antipsychotic medication.
In developing the models, medication was treated as a time-varying explanatory variable in order to accurately model exposure, as subjects’ medication statuses changed over time. When a medication was listed at a given time point, the exposure was assumed to have been started at the mid-point between the current and previous time points, and to have lasted until the mid-point between the current and subsequent time points. The mean change in Mini-Mental State Exam (MMSE) score over time was modeled using time slopes, with time-varying drug and covariate interactions as slope modifiers. Longitudinal regression models were developed using a random time slope by subject and a continuous 1st order auto-regressive covariance structure for errors on the same subjects, and were fit using MMSE scores extracted from the ‘NACCMMSE’ column, and using subjects’ age and sex, identified in the ‘NACCAGE’ and ‘SEX’ columns, as covariates. Only two-way interactions were considered and linear effects were assumed. Central limit theorems protect against non-severe departures from normality, and MMSE is a validated scale. APOE models were also developed using the presence or absence of an APOE4 allele as a modifier of the drug effect. The ‘NACCNE4S’ column was evaluated and subjects with a ‘1’ or a ‘2’ were designated APOE4 carriers, subjects with a ‘0’ were designated APOE4 non-carriers, and subjects with a ‘9’ (missing data) were excluded. Models were also developed where the baseline MMSE score, recorded at a subject’s first clinic visit, was included as a covariate. Linear combinations of parameters were tested with T and F tests, and the Satterthwaite method was used to calculate the denominator degrees of freedom. Model outputs were used in power analysis calculations performed in PASS 13 software (NCSS). All tests were two-sided, and 95% confidence intervals were presented for all univariate contrasts.
For reversion and conversion models, Cox proportional hazards models were developed, stratified by clinical diagnosis. The Cox model makes no parametric assumptions about the shape of the underlying hazard function, and stratification permits different underlying hazard functions for different clinical diagnoses. Tests for violation of the proportional hazards assumptions are not available for models with time-varying covariates. Clinical diagnoses were extracted from the ‘NACCUDSD’ column, in which a ‘1’ was considered “normal cognition (NC)”, ‘2’ [cognitively impaired, but not meeting the classical definition for mild cognitive impairment (MCI)] or ‘3’ were considered “MCI”, and ‘4’ was considered “AD”. A ‘4’ in the ‘NACCUDSD’ column indicates a diagnosis of dementia, which may include AD, Lewy body dementia, frontotemporal dementia, etc. However, the ‘NACCALZD’ column indicated that the vast majority of subjects receiving a dementia diagnosis were deemed to be of AD etiology (e.g., 29/32 subjects who took imipramine), and thus herein, we refer to this group collectively as AD patients. Drug exposure was modeled using time-varying covariates and cumulative exposure, controlled for time since last exposure, was selected for antidepressants, while on/off status was selected for antipsychotics. The variance calculations accounted for repeated measures, as the subjects could have multiple reversion/conversion events. Multiple reversion/conversion events were handled by aggregating observations within each subject and then the robust sandwich method was used for standard errors and tests. Subjects with an initial clinical diagnosis of AD were excluded from the risk set for conversion, and subjects with an initial clinical diagnosis of NC were excluded from the risk set for reversion, as they were ineligible for the event. Age and sex were controlled for, and in the interaction models, all two-way interactions between drug exposure, age, and sex were considered.
For the MMSE models, effects were assumed to be linear and only two-way interactions were considered. Linear combinations of parameters were tested with Z and χ2 tests. Hazard was modeled on a logarithmic scale and then the results were transformed back to hazard ratios. All tests were two-sided and 95% confidence intervals were presented for univariate contrasts. For the mixed medications models, subjects taking doxepin, citalopram, fluoxetine, aripiprazole, or quetiapine were identified by searching the ‘DRUGS’ column, and reversion models were developed as described above. Multiple testing adjustment was not applied because of the exploratory nature of the study and the complexity of the models; however, the conclusions drawn from these models would remain unchanged.
Statistical analyses
DOE and statistical analyses for the development of the fibrillization assay were performed using Minitab 18. Linear regression and one-way analysis of variance (ANOVA) were performed using GraphPad Prism 8. Following ANOVA, comparisons between multiple groups was done by post-hoc testing using the Holm-Šidák method and a P < 0.05 was considered statistically significant. All statistical tests were two-sided. Sample sizes, experimental replication, and exact statistical tests used are detailed in the figure legends. Except in the case of the kinetic Aβ fibrillization plots, where measurements were taken repetitively from the same wells, all measurements were taken from distinct samples.